Top-Down Tree Structured Text Generation

TL;DR

This paper proposes a top-down, tree-structured approach to text generation that explicitly models syntactic structures, improving the generation of complex, long sentences by performing global planning and fixing dependencies.

Contribution

It introduces a novel top-down, breadth-first tree generation method that explicitly models syntactic structures, contrasting with traditional sequential and depth-first approaches.

Findings

Effective on two generation tasks

Improves handling of complex sentence structures

Demonstrates advantages over transition-based methods

Abstract

Text generation is a fundamental building block in natural language processing tasks. Existing sequential models performs autoregression directly over the text sequence and have difficulty generating long sentences of complex structures. This paper advocates a simple approach that treats sentence generation as a tree-generation task. By explicitly modelling syntactic structures in a constituent syntactic tree and performing top-down, breadth-first tree generation, our model fixes dependencies appropriately and performs implicit global planning. This is in contrast to transition-based depth-first generation process, which has difficulty dealing with incomplete texts when parsing and also does not incorporate future contexts in planning. Our preliminary results on two generation tasks and one parsing task demonstrate that this is an effective strategy.